1. The Selection of Employment Discrimination Disputes for Litigation：using Business cycle effects to test the Priest-Klein Hypothesis
Peter Siegelman and John J.Donohue III
The Journal of Legal Studies 1995
報告人 ：高培儒

2. Abstract
In this paper, they confirm two predictions of Priest-Klein model of litigation.
Relatively weak cases should be more likely to settle.
Party with the greater stake in litigation will have the higher win rate in adjudicated disputes.
However, the strong version of P-K model predict a constant win rate over cycle, but the win rate falls during recession.
The observed settlement and win rate effects cannot be explained by change in relative stakes over business cycle, nor by variations over the cycle in the types of cases brought.

3. Introduction P-K model：
The process of negotiation tends to weed out the extremely strong cases through settlement, thereby pushing the plaintiff win rates toward 50 percent.
Party with more at stake will have higher win rate by simply offering more in settlement negotiation.

4. Introduction
In this article, instead of looking for a 50 percent win rate, they ask whether settlement is a random process or acts to screen out extremely weak or strong cases.
They use data on employment discrimination over 20 years, and demonstrate six points ：
The plaintiff win rate in employment discrimination cases is far below 50 percent, which may comes from that the defendant employers have a larger stake in litigation according to P-K model.

5. Introduction
A slumping economy tends to elicit a substantial increase in the number of employment discrimination cases filed.
These incremental (recession-induced) cases tend to be much weaker than average.
As P-K model predicts that the weaker cases will settle at a disproportionately high rate, in fact, cases filed during recession do have higher settlement rates.

6. Introduction
Contradict to strongest form of P-K model, plaintiff win rate for employment discrimination cases filed during recession has found to be lower than normal time.
Other explanations like changing in the composition of cases, in relative stakes, in the amount of discrimination, or in legal doctrine cannot explain their findings of significant business cycle effects on settlement and win rates.

7. Model
The P-K model：

8. Model
Model in this paper：

9. Model They use a slightly different model for two reason：
They want to capture the one-way fee-shifting rules.
Their test derives from differences in case quality over business cycle.
The main difference is that they allow intermediate stages between the initial dispute and the final adjudication.

10. Notation

11. Model
The expected value of the suit to the plaintiff at the time the file/drop decision is made is：
A1>0：a suit will be filed.
A1<0：a suit will not be filed.

12. Model
Since they allow the plaintiff to have access to new information and new costs at the settle/litigate decision, the lowest asking price of the plaintiff is：
The defendant’s maximum biding price is：
A2>B：litigation occurs.
A2<B：settlement reached.

13. Model
Their model makes three predictions about the effects of the business cycle on employment discrimination litigation.
Case filings will increase as the unemployment rate rise.
These additional recession-induced cases will likely be of weaker quality.
If they are in fact weaker than the average cases filed in the non-recessionary times, then we would expect to see a disproportionate number of them settle according to P-K model.

14. Prediction 1 Case filings：
Any case falling in the unshaded northeast region of the below figure will have a positive expected value for plaintiff and therefore meet A1>0.
When the unemployment rate rises, the duration of unemployment spells also increase, which prompts a proportionate increase in the damages that successful employment discrimination litigants will receive.
These will increase the number of cases filed.

15. Figure 2

16. Prediction 2
Most incremental cases are weaker, if the actual case distributed is uniform.

17. A testable model
Since they predict that higher unemployment leads to weaker cases that are disproportionately settled, a test of the selection model can thus be conducted by regressing a dichotomous variable on the unemployment rate at the filing time.(logistic model)
If the coefficients are insignificantly different from zero, we reject the selection model.
In addition, they have four hypotheses stemming from their model：

18. Hypotheses
Filing rate should be higher in slumps than in booms, and these incremental cases should on average be weaker.
Settlement rate for filed cases should be higher in slumps than in booms.
Observed win rate should be lower in slumps than in booms.
Sensitivity to the business cycle should increase as one moves down the disputing pyramid.

19. Data
The Administrative Office of the U.S. Courts (AO) did not begin keeping track of the statutory basis for individual suits until 1988.
They examine the grouping data quarterly instead of individual data.
The data are drawn from the data tapes of the Administrative Office of the U.S. Courts between July 1969 and June 1989.

20. TABLE 2

21. TABLE 3

22. TABLE 4

23. Interpret the result The result accepts all of the four hypotheses.
Table 5 presents the effect of a hypothetical increase in the rate of unemployment from 6.04 percent (the base level) to 8.66 percent.

24. Interpret the result
It shows that an average quarter with a 6.04 percent unemployment rate generates about 1,367 employment discrimination suits.
Raising the unemployment to 8.66 percent increases the number of cases by 515, to 1,882.
The settlement rate is 61.3 percent on the base cases, but 84.6 percent on the incremental cases.
A comparison of the 20.9 percent win rate for the base filings with the 16.4 percent win rate for the incremental cases confirms that the induced filings are in fact weaker .

25. Interpret the result
Thus, they have confirmed that a higher rate of unemployment induces a substantial number of additional cases, that these incremental cases are weaker than average, and thus have higher settlement rates and lower plaintiff win rates.
But the remaining incremental cases that have not settled are won by the plaintiffs at a rate of only 16.4 percent as opposed to the base plaintiff win rate of 20.9 percent.
So, they have shown that settlement does not completely filter out all of the additional low-quality cases as P-K model predict.

26. Some Alternative Explanations
They consider four other factors that might link the business cycle with the settlement and win rates：
Cyclical Changes in the Composition of the Caseload
Changing Doctrine
Changes in Discrimination
Changes in Relative Stakes

27. 1. Cyclical Changes in the Composition of the Caseload
If the entire composition of employment discrimination cases changes with the unemployment rate, then it does not properly identify the "incremental cases" that are added on to the constant "base filings."
The Worker Benefits Effect ：
The increase in filings during recessions comes from a rise in potential damages caused by the longer spells of unemployment suffered by most plaintiffs.
But employment discrimination case filings could increase in recessions if the number of incidents that give rise to litigation, such as discharges and layoffs, grew in recessions (the incidents effect ).

28. 1. Cyclical Changes in the Composition of the Caseload
This concern is alleviated by their earlier finding (in footnote 41) that the worker benefits effect, rather than the incidents effect, was the explanation for the countercyclical pattern.
Direct Tests for Cyclical Composition Changes：
Table 6 presents results from logistic regressions designed to test whether the business cycle influences the composition of the employment discrimination caseload.
It reveals that the business cycle effect is not statistically significant at conventional levels for any of the multinomial variables.

30. 1. Cyclical Changes in the Composition of the Caseload
The probability of a case containing a claim under $ 1983, and the probability of a sex discrimination claim, do seem to respond to the state of the economy.
Hiring and firing claims also appear somewhat sensitive to economic conditions.
In general, the results support that： the composition of cases does not change much over the cycle, which suggests that it is changes in case quality that drive the observed win rate and settlement rate effects.

31. 2. Changing Doctrine
Whether or not there has been a long-term change in doctrine that might explain the falling win rate, it is highly implausible that there is any relationship between changes in legal doctrine and the business cycle.
Moreover, even if one assumed that courts were more sympathetic to plaintiffs who file during economic downturns, this assumption would be consistent with the higher rate of filings in downturns, but not with the lower win rate.

32. 3. Changes in Discrimination
Since unemployment represents an excess supply of labor at the going wage, it presents employers have an opportunity to pick and choose among workers queuing for jobs.
Then, it suggests, that discrimination should rise during slumps, and this might be expected to induce more case filings.
But if discrimination is indeed higher during slumps, then the incremental cases induced by this greater discrimination should tend to be stronger than the average base case (or at least no weaker).
In this event, the obvious prediction is that plaintiffs should win more (or at least, no fewer) of the incremental cases that are filed at such times.
This is contradict with their finding.

33. 4. Changes in Relative Stakes
Defining the ratio of the plaintiff's to the defendant's stakes as R.
Since the selection model predicts that the party with the larger stakes will prevail more often, the results presented earlier (in Tables 1-4) require that：
R is less than unity
R is decreasing over time (since the win rate is falling)
R moves pro-cyclically. (since plaintiffs win a larger share of the cases filed in a booming economy, R must be greater in booms than in slumps to be consistent with the theory)

34. 4. Changes in Relative Stakes
Condition (1)：defendants may have reputation and adverse-precedent effects from losing a lawsuit which would not be matched by symmetric gains to plaintiffs.
condition (2)：if the stigma of being labeled a discriminator has increased over time, relative stakes will be decreasing.
condition (3)：relative stakes should tend to move counter-cyclically, in violation of condition (3)

35. 4. Changes in Relative Stakes
Suppose that the plaintiff's stakes are limited to her back pay award, J.
The defendant's stake in the litigation consists of the back pay award, J , plus reputation or precedential effects, denote by K, K > 0.
This means that relative stakes (R) increase during a slump, rather than decreasing as the selection model requires in order to explain the falling win rate.

36. Conclusion
This article shows how the relationship between business cycle, as measured by unemployment rate, and all phases of employment discrimination litigation.
The influence of the business cycle can be detected in the plaintiff win rate, implying only partial selection in the settlement of employment discrimination disputes, rather than perfect selection as strict P-K model.
They confirms that weak cases tend to settle at a disproportionately high rate, consistently with P-K model.